Since LinkedIn added the "Seems like AI slop" button in July 2026, one question comes up more than any other: will my posts get flagged? It usually comes from people who do use AI to draft, and who now feel vaguely guilty about it.
Here is the reassuring part. LinkedIn has said plainly that it does not penalise AI use, and the authenticity flag in your dashboard runs on reports from readers rather than a detector. Nothing is scanning your posts to work out which tool wrote them. If a post gets reported, a person read it and decided it was not worth their time.
So the useful question is not whether a machine will catch you. It is whether a reader would bother reporting you. Below is how to check that, in the order worth checking it.
1. Is there anything in it a reader could check?
Read your draft and count the particulars. Numbers, dates, names of people, tools, companies, places. Count the moments where something changed, with a before and an after.
If the count is zero, stop there. Nothing else on this list will save the post. This is the one thing a model genuinely cannot fix for you, because it was never given the information in the first place. Ask a language model to write about leadership and it returns the average of every leadership post ever written, because the average is all it has.
The test: could a competitor in your industry publish this exact post under their own name and have it fit? If yes, you have written the industry's post rather than yours.
2. Does the shape give it away before the words do?
Readers pattern-match on layout faster than they read. Four shapes get discounted on sight, and all four are visible from the white space alone, before a single word registers.
- A short standalone opening line above a blank line. The single most recognisable template on the platform.
- One sentence per paragraph, all the way down. Sometimes called broetry. It was an engagement hack years ago and now reads as a tell.
- Exactly three of everything. Three lessons, three mistakes, three adjectives. Real counts are rarely three.
- A closing line engineered to be quotable. If it would work on a poster, it is doing decoration rather than work.
Shape is the tell that travels fastest. Your reader is moving at scrolling speed, and the silhouette of a post arrives whole, in a glance, while the words are still queuing up behind it.
3. Could anyone disagree with you?
Find the sentence in your post that someone could argue with. If there is no such sentence, you have written coverage rather than a position.
Balanced, safe, on-the-one-hand writing is what models produce by default, because they are trained to avoid taking sides. It is also what readers skim. A post that names a cost, admits a thing that did not work, or says "I think most people have this backwards" gives the reader something to hold on to.
You do not need to pick a fight. You need one sentence that could be wrong.
4. Does the reader learn anything after the first line?
Cover everything except your opening sentence. Can a reader predict the rest? If your first line is the whole point and the body restates it in fresh words three times over, there is no post underneath.
The fix is usually to move the interesting part up. The thing you almost cut because it felt too specific is normally the reason to publish at all.
5. Only now, check the words
Vocabulary is last on this list on purpose. It is the easiest thing to fix, the easiest thing to check, and the one thing every writing tool on the market already does for you. The four checks above it are the ones no tool can do on your behalf.
That said, the surface tells are real and free to remove.
| Tell | Example |
|---|---|
| Hype vocabulary | leverage, unlock, empower, elevate, seamless, game-changer, delve |
| Negative parallelism | "It is not just X, it is Y" |
| Trailing analysis | "..., highlighting how much this matters" |
| Em dashes and curly quotes | Common in model output, rare in phone-typed posts |
| Engagement bait close | "Agree?", "Thoughts?", "Follow me for more" |
Why a banned-word list is not enough
A word list reaches exactly one of the five things above. It can strip every "leverage" out of your draft and it will never once ask whether the post says anything, what shape it arrived in, or whether you took a position. Those live in the post as a whole, not in any single word of it.
There is a reason word lists are the industry default anyway: they are the easy half. Word choice is the one instruction a language model follows perfectly on the first ask. Ban "delve" and it will never write "delve" again. The template underneath comes back untouched in the next draft, wearing different vocabulary.
LinkedIn reached the same conclusion when it designed the button. It could have built an origin detector, and it deliberately did not. Peer-reviewed evaluations have found detectors are neither accurate nor reliable, and a Stanford study found they wrongly flagged 61% of TOEFL essays by non-native English speakers as machine-written. So LinkedIn built the flag on what readers report instead, which makes a reader's reaction the actual standard your post is measured against.
Which is the practical lesson, and it holds no matter what you draft with. A clean word count is the floor, not the finish line. Check the four structural things first, and treat vocabulary as the last mile.
Check a draft in about ten seconds
Our free AI slop checker runs all five checks for you, quotes back the line a reader is most likely to bounce off, and asks the one question that would make the post specific. No login, and it works on anything you paste, whether you wrote it yourself, in ChatGPT, or in FeedBoss.
What it will not do is tell you whether AI wrote something. It also cannot promise your post will never be reported, because that button is pressed by people and people disagree with each other. What it can tell you is where a reader is most likely to lose interest, and which line to fix first.
For the background, see what AI slop is and how to humanize AI writing. For how a generator can handle these five checks before you ever see the draft, see how FeedBoss prevents AI slop.
FAQs
Will LinkedIn flag my post if I used AI to write it?
Not for that reason. LinkedIn has said it does not penalise AI use and that the authenticity flag runs on reader reports rather than an automated detector. What gets reported is writing that reads as generic, whatever produced it.
What happens if someone reports my post as AI slop?
The post is hidden from that person's feed, the report feeds LinkedIn's classifier training, and the post's reach outside your network is reduced. You may also see a private note in your own analytics that readers are finding your content inauthentic.
How can I tell if my own post reads as AI?
Check five things in order: whether it contains anything a reader could verify, whether its shape follows a template, whether anyone could disagree with it, whether the body adds anything beyond the first line, and only then the vocabulary. The first four are what a reader reacts to, and a banned-word list reaches none of them.
Are AI detectors accurate enough to rely on?
No. Peer-reviewed evaluations have found detectors are neither accurate nor reliable, and one Stanford study found detectors classified the majority of essays by non-native English speakers as AI-generated. That is why LinkedIn's own authenticity signal uses reader reports instead.